generic-library/ffmpeg
1
0
Fork 0
Code Issues Pull Requests Releases Wiki Activity

3dnow dering

Browse Source 
Originally committed as revision 2476 to svn://svn.mplayerhq.hu/mplayer/trunk/postproc
This commit is contained in:
Michael Niedermayer 2001-10-25 22:50:03 +00:00
parent cf76316aa3
commit e0f8ffae99
2 changed files with 64 additions and 20 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

42
postproc/postprocess.c
View File

@ -26,7 +26,7 @@ isHorizDC Ec Ec
isHorizMinMaxOk a E isHorizMinMaxOk a E
doHorizLowPass E e e doHorizLowPass E e e
doHorizDefFilter Ec Ec Ec doHorizDefFilter Ec Ec Ec
deRing e deRing e e*
Vertical RKAlgo1 E a a Vertical RKAlgo1 E a a
Horizontal RKAlgo1 a a Horizontal RKAlgo1 a a
Vertical X1 a E E Vertical X1 a E E
@ -1808,7 +1808,7 @@ Implemented Exact 7-Tap
static inline void dering(uint8_t src[], int stride, int QP) static inline void dering(uint8_t src[], int stride, int QP)
{ {
#ifdef HAVE_MMX2 #if defined (HAVE_MMX2) || defined (HAVE_3DNOW)
asm volatile( asm volatile(
"movq pQPb, %%mm0 \n\t" "movq pQPb, %%mm0 \n\t"
"paddusb %%mm0, %%mm0 \n\t" "paddusb %%mm0, %%mm0 \n\t"
@ -1821,11 +1821,20 @@ static inline void dering(uint8_t src[], int stride, int QP)
"pcmpeqb %%mm6, %%mm6 \n\t" "pcmpeqb %%mm6, %%mm6 \n\t"
"pxor %%mm7, %%mm7 \n\t" "pxor %%mm7, %%mm7 \n\t"
#ifdef HAVE_MMX2
#define FIND_MIN_MAX(addr)\ #define FIND_MIN_MAX(addr)\
"movq " #addr ", %%mm0 \n\t"\ "movq " #addr ", %%mm0 \n\t"\
"pminub %%mm0, %%mm6 \n\t"\ "pminub %%mm0, %%mm6 \n\t"\
"pmaxub %%mm0, %%mm7 \n\t" "pmaxub %%mm0, %%mm7 \n\t"
#else
#define FIND_MIN_MAX(addr)\
"movq " #addr ", %%mm0 \n\t"\
"movq %%mm6, %%mm1 \n\t"\
"psubusb %%mm0, %%mm7 \n\t"\
"paddb %%mm0, %%mm7 \n\t"\
"psubusb %%mm0, %%mm1 \n\t"\
"psubb %%mm1, %%mm6 \n\t"
#endif
FIND_MIN_MAX((%%eax)) FIND_MIN_MAX((%%eax))
FIND_MIN_MAX((%%eax, %1)) FIND_MIN_MAX((%%eax, %1))
@ -1838,36 +1847,49 @@ FIND_MIN_MAX((%0, %1, 8))
"movq %%mm6, %%mm4 \n\t" "movq %%mm6, %%mm4 \n\t"
"psrlq $8, %%mm6 \n\t" "psrlq $8, %%mm6 \n\t"
"pminub %%mm4, %%mm6 \n\t" // min of pixels
#ifdef HAVE_MMX2 #ifdef HAVE_MMX2
"pminub %%mm4, %%mm6 \n\t" // min of pixels
"pshufw $0xF9, %%mm6, %%mm4 \n\t" "pshufw $0xF9, %%mm6, %%mm4 \n\t"
"pminub %%mm4, %%mm6 \n\t" // min of pixels "pminub %%mm4, %%mm6 \n\t" // min of pixels
"pshufw $0xFE, %%mm6, %%mm4 \n\t" "pshufw $0xFE, %%mm6, %%mm4 \n\t"
"pminub %%mm4, %%mm6 \n\t"
#else #else
"movq %%mm6, %%mm1 \n\t"
"psubusb %%mm4, %%mm1 \n\t"
"psubb %%mm1, %%mm6 \n\t"
"movq %%mm6, %%mm4 \n\t" "movq %%mm6, %%mm4 \n\t"
"psrlq $16, %%mm6 \n\t" "psrlq $16, %%mm6 \n\t"
"pminub %%mm4, %%mm6 \n\t" "movq %%mm6, %%mm1 \n\t"
"psubusb %%mm4, %%mm1 \n\t"
"psubb %%mm1, %%mm6 \n\t"
"movq %%mm6, %%mm4 \n\t" "movq %%mm6, %%mm4 \n\t"
"psrlq $32, %%mm6 \n\t" "psrlq $32, %%mm6 \n\t"
"movq %%mm6, %%mm1 \n\t"
"psubusb %%mm4, %%mm1 \n\t"
"psubb %%mm1, %%mm6 \n\t"
#endif #endif
"pminub %%mm4, %%mm6 \n\t"
"movq %%mm7, %%mm4 \n\t" "movq %%mm7, %%mm4 \n\t"
"psrlq $8, %%mm7 \n\t" "psrlq $8, %%mm7 \n\t"
"pmaxub %%mm4, %%mm7 \n\t" // max of pixels
#ifdef HAVE_MMX2 #ifdef HAVE_MMX2
"pmaxub %%mm4, %%mm7 \n\t" // max of pixels
"pshufw $0xF9, %%mm7, %%mm4 \n\t" "pshufw $0xF9, %%mm7, %%mm4 \n\t"
"pmaxub %%mm4, %%mm7 \n\t" // min of pixels "pmaxub %%mm4, %%mm7 \n\t"
"pshufw $0xFE, %%mm7, %%mm4 \n\t" "pshufw $0xFE, %%mm7, %%mm4 \n\t"
"pmaxub %%mm4, %%mm7 \n\t"
#else #else
"psubusb %%mm4, %%mm7 \n\t"
"paddb %%mm4, %%mm7 \n\t"
"movq %%mm7, %%mm4 \n\t" "movq %%mm7, %%mm4 \n\t"
"psrlq $16, %%mm7 \n\t" "psrlq $16, %%mm7 \n\t"
"pmaxub %%mm4, %%mm7 \n\t" "psubusb %%mm4, %%mm7 \n\t"
"paddb %%mm4, %%mm7 \n\t"
"movq %%mm7, %%mm4 \n\t" "movq %%mm7, %%mm4 \n\t"
"psrlq $32, %%mm7 \n\t" "psrlq $32, %%mm7 \n\t"
"psubusb %%mm4, %%mm7 \n\t"
"paddb %%mm4, %%mm7 \n\t"
#endif #endif
"pmaxub %%mm4, %%mm7 \n\t"
PAVGB(%%mm6, %%mm7) // a=(max + min)/2 PAVGB(%%mm6, %%mm7) // a=(max + min)/2
"punpcklbw %%mm7, %%mm7 \n\t" "punpcklbw %%mm7, %%mm7 \n\t"
"punpcklbw %%mm7, %%mm7 \n\t" "punpcklbw %%mm7, %%mm7 \n\t"

42
postproc/postprocess_template.c
View File

@ -26,7 +26,7 @@ isHorizDC Ec Ec
isHorizMinMaxOk a E isHorizMinMaxOk a E
doHorizLowPass E e e doHorizLowPass E e e
doHorizDefFilter Ec Ec Ec doHorizDefFilter Ec Ec Ec
deRing e deRing e e*
Vertical RKAlgo1 E a a Vertical RKAlgo1 E a a
Horizontal RKAlgo1 a a Horizontal RKAlgo1 a a
Vertical X1 a E E Vertical X1 a E E
@ -1808,7 +1808,7 @@ Implemented Exact 7-Tap
static inline void dering(uint8_t src[], int stride, int QP) static inline void dering(uint8_t src[], int stride, int QP)
{ {
#ifdef HAVE_MMX2 #if defined (HAVE_MMX2) || defined (HAVE_3DNOW)
asm volatile( asm volatile(
"movq pQPb, %%mm0 \n\t" "movq pQPb, %%mm0 \n\t"
"paddusb %%mm0, %%mm0 \n\t" "paddusb %%mm0, %%mm0 \n\t"
@ -1821,11 +1821,20 @@ static inline void dering(uint8_t src[], int stride, int QP)
"pcmpeqb %%mm6, %%mm6 \n\t" "pcmpeqb %%mm6, %%mm6 \n\t"
"pxor %%mm7, %%mm7 \n\t" "pxor %%mm7, %%mm7 \n\t"
#ifdef HAVE_MMX2
#define FIND_MIN_MAX(addr)\ #define FIND_MIN_MAX(addr)\
"movq " #addr ", %%mm0 \n\t"\ "movq " #addr ", %%mm0 \n\t"\
"pminub %%mm0, %%mm6 \n\t"\ "pminub %%mm0, %%mm6 \n\t"\
"pmaxub %%mm0, %%mm7 \n\t" "pmaxub %%mm0, %%mm7 \n\t"
#else
#define FIND_MIN_MAX(addr)\
"movq " #addr ", %%mm0 \n\t"\
"movq %%mm6, %%mm1 \n\t"\
"psubusb %%mm0, %%mm7 \n\t"\
"paddb %%mm0, %%mm7 \n\t"\
"psubusb %%mm0, %%mm1 \n\t"\
"psubb %%mm1, %%mm6 \n\t"
#endif
FIND_MIN_MAX((%%eax)) FIND_MIN_MAX((%%eax))
FIND_MIN_MAX((%%eax, %1)) FIND_MIN_MAX((%%eax, %1))
@ -1838,36 +1847,49 @@ FIND_MIN_MAX((%0, %1, 8))
"movq %%mm6, %%mm4 \n\t" "movq %%mm6, %%mm4 \n\t"
"psrlq $8, %%mm6 \n\t" "psrlq $8, %%mm6 \n\t"
"pminub %%mm4, %%mm6 \n\t" // min of pixels
#ifdef HAVE_MMX2 #ifdef HAVE_MMX2
"pminub %%mm4, %%mm6 \n\t" // min of pixels
"pshufw $0xF9, %%mm6, %%mm4 \n\t" "pshufw $0xF9, %%mm6, %%mm4 \n\t"
"pminub %%mm4, %%mm6 \n\t" // min of pixels "pminub %%mm4, %%mm6 \n\t" // min of pixels
"pshufw $0xFE, %%mm6, %%mm4 \n\t" "pshufw $0xFE, %%mm6, %%mm4 \n\t"
"pminub %%mm4, %%mm6 \n\t"
#else #else
"movq %%mm6, %%mm1 \n\t"
"psubusb %%mm4, %%mm1 \n\t"
"psubb %%mm1, %%mm6 \n\t"
"movq %%mm6, %%mm4 \n\t" "movq %%mm6, %%mm4 \n\t"
"psrlq $16, %%mm6 \n\t" "psrlq $16, %%mm6 \n\t"
"pminub %%mm4, %%mm6 \n\t" "movq %%mm6, %%mm1 \n\t"
"psubusb %%mm4, %%mm1 \n\t"
"psubb %%mm1, %%mm6 \n\t"
"movq %%mm6, %%mm4 \n\t" "movq %%mm6, %%mm4 \n\t"
"psrlq $32, %%mm6 \n\t" "psrlq $32, %%mm6 \n\t"
"movq %%mm6, %%mm1 \n\t"
"psubusb %%mm4, %%mm1 \n\t"
"psubb %%mm1, %%mm6 \n\t"
#endif #endif
"pminub %%mm4, %%mm6 \n\t"
"movq %%mm7, %%mm4 \n\t" "movq %%mm7, %%mm4 \n\t"
"psrlq $8, %%mm7 \n\t" "psrlq $8, %%mm7 \n\t"
"pmaxub %%mm4, %%mm7 \n\t" // max of pixels
#ifdef HAVE_MMX2 #ifdef HAVE_MMX2
"pmaxub %%mm4, %%mm7 \n\t" // max of pixels
"pshufw $0xF9, %%mm7, %%mm4 \n\t" "pshufw $0xF9, %%mm7, %%mm4 \n\t"
"pmaxub %%mm4, %%mm7 \n\t" // min of pixels "pmaxub %%mm4, %%mm7 \n\t"
"pshufw $0xFE, %%mm7, %%mm4 \n\t" "pshufw $0xFE, %%mm7, %%mm4 \n\t"
"pmaxub %%mm4, %%mm7 \n\t"
#else #else
"psubusb %%mm4, %%mm7 \n\t"
"paddb %%mm4, %%mm7 \n\t"
"movq %%mm7, %%mm4 \n\t" "movq %%mm7, %%mm4 \n\t"
"psrlq $16, %%mm7 \n\t" "psrlq $16, %%mm7 \n\t"
"pmaxub %%mm4, %%mm7 \n\t" "psubusb %%mm4, %%mm7 \n\t"
"paddb %%mm4, %%mm7 \n\t"
"movq %%mm7, %%mm4 \n\t" "movq %%mm7, %%mm4 \n\t"
"psrlq $32, %%mm7 \n\t" "psrlq $32, %%mm7 \n\t"
"psubusb %%mm4, %%mm7 \n\t"
"paddb %%mm4, %%mm7 \n\t"
#endif #endif
"pmaxub %%mm4, %%mm7 \n\t"
PAVGB(%%mm6, %%mm7) // a=(max + min)/2 PAVGB(%%mm6, %%mm7) // a=(max + min)/2
"punpcklbw %%mm7, %%mm7 \n\t" "punpcklbw %%mm7, %%mm7 \n\t"
"punpcklbw %%mm7, %%mm7 \n\t" "punpcklbw %%mm7, %%mm7 \n\t"